Improved initialization of embedding and restriction: need to evaluate a tensor produ...

Improved initialization of embedding and restriction: need to evaluate a tensor product polynomial a lot of times on tensor product points. Hence, do the polynomial evaluation in one dimension at a time. This makes code slightly more complicated to read, but it evaluates much faster, especially for high orders. Now, FE_Q can be initialized for order 15 in 3D in less than a second. Before it took hundreds of seconds.

git-svn-id: https://svn.dealii.org/trunk@23961 0785d39b-7218-0410-832d-ea1e28bc413d